Nutrient Document Engine MCP Server

import { DocumentEngineClient } from '../../api/Client.js'; import { z } from 'zod'; import { DocumentFingerprintSchema } from '../schemas/DocumentFingerprintSchema.js'; import { MCPToolOutput } from '../../mcpTools.js'; import { getDocumentInfo } from '../../api/DocumentLayerAbstraction.js'; /** * Schema for extract_tables tool */ // Create a Zod schema for PageRange const PageRangeSchema = z.object({ start: z.number().optional(), end: z.number().optional(), }); export const ExtractTablesSchema = { document_fingerprint: DocumentFingerprintSchema, page_range: PageRangeSchema.optional().describe( 'Range of pages to include with start and end indices (0-based)' ), }; export const ExtractTablesInputSchema = z.object(ExtractTablesSchema); export type ExtractTablesInput = z.infer<typeof ExtractTablesInputSchema>; /** * Extract tables from a document using the build-document API with JSONContentOutput */ export async function extractTables( client: DocumentEngineClient, params: ExtractTablesInput ): Promise<MCPToolOutput> { try { // Validate input const validatedParams = ExtractTablesInputSchema.parse(params); const { document_fingerprint, page_range } = validatedParams; // Get document info to get total page count and title const docInfo = await getDocumentInfo(client, document_fingerprint); const pageCount = docInfo.pageCount || 0; const title = docInfo.title || 'Untitled Document'; // Prepare document part with optional page range const documentPart: { document: { id: string; layer?: string }; pages?: { start?: number; end?: number }; } = { document: { id: document_fingerprint.document_id, layer: document_fingerprint.layer, }, }; // Add page range if provided if (page_range) { documentPart.pages = page_range; } // Build request with JSONContentOutput for tables const response = await client['build-document']( {}, { parts: [documentPart], output: { type: 'json-content', plainText: false, structuredText: false, keyValuePairs: false, tables: true, }, } ); // Extract the tables from the response // Ensure pages exist in the response const pages = response.data.pages || []; const tables = pages.flatMap((page, pageIndex) => { return (page.tables || []).map(table => ({ ...table, pageIndex, })); }); // Build the markdown content let markdown = `# Table Extraction\n\n`; markdown += `📄 **Document:** ${title} \n`; markdown += `📄 **Document ID:** ${document_fingerprint.document_id} \n`; if (document_fingerprint.layer) { markdown += `🔀 **Layer:** ${document_fingerprint.layer} \n`; } markdown += `📑 **Total Pages:** ${pageCount} \n`; // Add page range information if provided if (page_range) { let pageRangeText = 'Pages '; if (page_range.start !== undefined) { pageRangeText += `from ${page_range.start}`; } if (page_range.end !== undefined) { pageRangeText += `${page_range.start !== undefined ? ' to ' : ''}${page_range.end}`; } markdown += `📖 **Pages Processed:** ${pageRangeText} \n`; } else { markdown += `📖 **Pages Processed:** All pages \n`; } markdown += `📋 **Tables Found:** ${tables.length} \n\n`; markdown += `---\n\n`; if (tables.length === 0) { markdown += `## No Tables Found\n\n`; markdown += `No tables were detected in the document. This could be because:\n\n`; markdown += `- The document doesn't contain any tables\n`; markdown += `- The tables are represented as images and not as structured data\n`; markdown += `- The table detection algorithm couldn't recognize the tables\n\n`; } else { markdown += `## Extracted Tables\n\n`; tables.forEach((table, tableIndex) => { const pageNumber = table.pageIndex + 1; // Convert to 1-based for display markdown += `### Table ${tableIndex + 1} (Page ${pageNumber})\n\n`; // Create markdown table if (table.cells && table.cells.length > 0) { // Group cells by row const rowsMap = new Map(); table.cells.forEach(cell => { if (!rowsMap.has(cell.rowIndex)) { rowsMap.set(cell.rowIndex, []); } rowsMap.get(cell.rowIndex).push(cell); }); // Sort rows by index const rows = Array.from(rowsMap.entries()) .sort(([rowIndexA], [rowIndexB]) => rowIndexA - rowIndexB) .map(([_, cells]) => cells); // Find the maximum column index to determine table width const maxColIndex = Math.max(...table.cells.map(cell => cell.columnIndex)); // Create header row markdown += '| '; for (let i = 0; i <= maxColIndex; i++) { markdown += `Column ${i + 1} | `; } markdown += '\n'; // Create separator row markdown += '| '; for (let i = 0; i <= maxColIndex; i++) { markdown += '--- | '; } markdown += '\n'; // Create data rows rows.forEach(cells => { markdown += '| '; // Sort cells by column index const sortedCells = [...cells].sort((a, b) => a.columnIndex - b.columnIndex); // Fill in empty cells let currentColIndex = 0; sortedCells.forEach(cell => { // Fill in any missing columns before this cell while (currentColIndex < cell.columnIndex) { markdown += ' | '; currentColIndex++; } // Add this cell's content markdown += `${cell.text || ''} | `; currentColIndex++; }); // Fill in any remaining columns while (currentColIndex <= maxColIndex) { markdown += ' | '; currentColIndex++; } markdown += '\n'; }); markdown += '\n'; } else { markdown += `*Table structure detected but no cells were found*\n\n`; } }); } return { markdown }; } catch (error) { // Provide a more user-friendly error message return { markdown: `# Error Extracting Tables\n\nAn error occurred while trying to extract tables: ${error instanceof Error ? error.message : String(error)}\n\nPlease check your connection and try again.`, }; } }